Hierarchical codebook for moving microbe detection in sewage

In this paper, we propose a newly object detection method for moving microbe in sewage. Firstly, two codebooks with YCrCb color model and improved local direction pattern are introduced in spatial context. Then, we design an adaptive update mechanism to accelerate convergence and enhance the accuracy of background based on self-learning strategy. Finally, we construct spatial and temporal relations in a MRF-MAP framework to further eliminate the left background pixels in foreground. Experiments results have showed the effectiveness and stability of our proposed method, compared with other algorithms.